import os

%matplotlib inline
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import sklearn
sklearn.set_config(print_changed_only=True)
from sklearn.model_selection import train_test_split, cross_val_score
from sklearn.pipeline import make_pipeline
from sklearn.preprocessing import scale, StandardScaler
import warnings
warnings.filterwarnings('ignore', category=DeprecationWarning)
warnings.filterwarnings('ignore', category=FutureWarning)

data = pd.read_csv('data_norm2.csv')
data

	agegroup	AL1	EducationG	eye	filter_$	Height	p1_digitalscreenH	p1_digitalscreenTVH	p1_outdoorH	p1_readingwritingH	...	agegroup2	AgeY	SEg	Sex	TVdistance_m	myopiabaselinegroups	Basline_AL_Binned	SEgroupsbaselinenew	risksMLV	risksSEP
0	6.0	22.39	0.0	1.0	1	105.0	0.50	1.50	0.00	0.88	...	1.0	6.90	0.0	2	4	0.0	1.0	0.0	0	0
1	6.0	22.41	0.0	2.0	1	105.0	0.50	1.50	0.00	0.88	...	1.0	6.90	1.0	2	4	1.0	1.0	1.0	0	0
2	6.0	22.88	1.0	2.0	1	108.1	1.56	2.99	2.93	1.07	...	1.0	6.84	0.0	1	3	0.0	2.0	0.0	0	0
3	6.0	22.94	1.0	1.0	1	108.1	1.56	2.99	2.93	1.07	...	1.0	6.84	0.0	1	3	0.0	2.0	0.0	0	0
4	6.0	22.83	0.0	2.0	1	108.2	2.31	2.31	1.79	1.07	...	1.0	6.43	1.0	1	2	1.0	2.0	1.0	0	0
...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
7436	12.0	23.46	1.0	1.0	1	171.4	0.21	1.64	1.43	1.00	...	3.0	12.64	1.0	2	2	1.0	3.0	1.0	0	0
7437	12.0	23.65	0.0	2.0	1	171.5	2.00	3.00	6.21	0.86	...	3.0	12.38	1.0	1	1	1.0	4.0	1.0	0	0
7438	12.0	23.75	0.0	1.0	1	171.5	2.00	3.00	6.21	0.86	...	3.0	12.38	1.0	1	1	1.0	4.0	1.0	0	0
7439	12.0	23.63	1.0	2.0	1	172.3	2.57	4.57	1.99	0.89	...	3.0	12.78	1.0	1	3	1.0	4.0	1.0	0	0
7440	12.0	23.53	1.0	1.0	1	172.3	2.57	4.57	1.99	0.89	...	3.0	12.78	1.0	1	3	1.0	4.0	1.0	0	0

7441 rows × 27 columns

target = data['risksSEP']
#target = data['risksMLV']
target.value_counts()

0    7277
1     164
Name: risksSEP, dtype: int64

X = data.iloc[:,:-2]
#X = data.dropna(subset=['AL_quartiles', 'Myopialastcat'], inplace=True)
X.shape

(7441, 25)

y=target #decision class
y.value_counts()

0    7277
1     164
Name: risksSEP, dtype: int64

X_train, X_test, y_train, y_test = train_test_split(X.values, y.values, random_state=0)

BASELINE imbalanced data

from imblearn.metrics import geometric_mean_score
from sklearn.model_selection import cross_validate
from sklearn.linear_model import LogisticRegression
from sklearn.metrics import make_scorer


gm_scorer = make_scorer(geometric_mean_score, greater_is_better=True)
scores = cross_validate(LogisticRegression(max_iter=5000),
                        X_train, y_train, cv=5, scoring={'roc_auc':'roc_auc', 'average_precision':'average_precision','accuracy':'accuracy','f1':'f1','gmean':gm_scorer})

print('Test roc: ',scores['test_roc_auc'].mean()) 
print('Test accuracy: ', scores['test_accuracy'].mean())
print('Test f1: ', scores['test_f1'].mean())
print('Test g mean: ', scores['test_gmean'].mean())

Test roc:  0.8767872909893276
Test accuracy:  0.9781362007168457
Test f1:  0.13317972350230414
Test g mean:  0.25056932160234285

model = LogisticRegression(max_iter=8000)
model.fit(X_train,y_train)
prediction = model.predict(X_test)

from sklearn.metrics import confusion_matrix
import seaborn as sns

cf_matrix = cf_matrix = confusion_matrix(y_test,prediction)
sns.heatmap(cf_matrix/np.sum(cf_matrix), annot=True, fmt='.2%', cmap='Pastel1')

<matplotlib.axes._subplots.AxesSubplot at 0x7f4dc0813f10>

from sklearn.metrics import ConfusionMatrixDisplay

ConfusionMatrixDisplay.from_predictions(y_test,prediction,cmap='viridis')

<sklearn.metrics._plot.confusion_matrix.ConfusionMatrixDisplay at 0x7f4dc096b790>

from sklearn.metrics import RocCurveDisplay

RocCurveDisplay.from_predictions(y_test,prediction)

<sklearn.metrics._plot.roc_curve.RocCurveDisplay at 0x7f4dc0b38090>

from sklearn.metrics import roc_auc_score
from sklearn.metrics import accuracy_score
from sklearn.metrics import f1_score
from imblearn.metrics import geometric_mean_score

print('Roc score: ',roc_auc_score(y_test,prediction)) 
print('Accuracy score: ',  accuracy_score(y_test,prediction))
print('f1 score: ',f1_score(y_test,prediction))
print('G mean score: ',  geometric_mean_score(y_test,prediction))

Roc score:  0.5216715614292707
Accuracy score:  0.9758194519075766
f1 score:  0.0816326530612245
G mean score:  0.21070238920905554

UNDERSAMPLE

from imblearn.pipeline import make_pipeline as make_imb_pipeline
from imblearn.under_sampling import RandomUnderSampler
undersample_pipe = make_imb_pipeline(RandomUnderSampler(random_state=42) ,LogisticRegression(max_iter=5000))
scores = cross_validate(undersample_pipe,
                        X_train, y_train, cv=5, scoring={'roc_auc':'roc_auc', 'average_precision':'average_precision','accuracy':'accuracy','f1':'f1','gmean':gm_scorer})
print('Test roc: ',scores['test_roc_auc'].mean()) 
print('Test accuracy: ', scores['test_accuracy'].mean())
print('Test f1: ', scores['test_f1'].mean())
print('Test g mean: ', scores['test_gmean'].mean())

Test roc:  0.8814577988508434
Test accuracy:  0.7919354838709677
Test f1:  0.14639953009792572
Test g mean:  0.8111251706623477

from imblearn.pipeline import make_pipeline as make_imb_pipeline

undersample_pipe.fit(X_train, y_train)
prediction = undersample_pipe.predict(X_test)
y_test.shape

(1861,)

from sklearn.metrics import confusion_matrix
import seaborn as sns

cf_matrix = cf_matrix = confusion_matrix(y_test,prediction)
sns.heatmap(cf_matrix/np.sum(cf_matrix), annot=True, fmt='.2%', cmap='Pastel1')

<matplotlib.axes._subplots.AxesSubplot at 0x7f4dc06a5450>

from sklearn.metrics import ConfusionMatrixDisplay

ConfusionMatrixDisplay.from_predictions(y_test,prediction,cmap='viridis')

<sklearn.metrics._plot.confusion_matrix.ConfusionMatrixDisplay at 0x7f4dc0634b50>

from sklearn.metrics import RocCurveDisplay

RocCurveDisplay.from_predictions(y_test,prediction)

<sklearn.metrics._plot.roc_curve.RocCurveDisplay at 0x7f4dc0548e50>

from sklearn.metrics import roc_auc_score
from sklearn.metrics import accuracy_score
from sklearn.metrics import f1_score
from imblearn.metrics import geometric_mean_score
print('Roc score: ',roc_auc_score(y_test,prediction)) 
print('Accuracy score: ',  accuracy_score(y_test,prediction))
print('f1 score: ',f1_score(y_test,prediction))
print('G mean score: ',  geometric_mean_score(y_test,prediction))

Roc score:  0.8346671561429271
Accuracy score:  0.8253627082213864
f1 score:  0.18952618453865333
G mean score:  0.8346098886169456

OVERSAMPLE

from imblearn.over_sampling import SMOTE
from imblearn.pipeline import make_pipeline as make_imb_pipeline
oversample_pipe = make_imb_pipeline(SMOTE(random_state=42), LogisticRegression(max_iter=5000))
scores = cross_validate(oversample_pipe,
                        X_train, y_train, cv=5, scoring={'roc_auc':'roc_auc', 'average_precision':'average_precision','accuracy':'accuracy','f1':'f1','gmean':gm_scorer})
print('Test roc: ',scores['test_roc_auc'].mean()) 
print('Test accuracy: ', scores['test_accuracy'].mean())
print('Test f1: ', scores['test_f1'].mean())
print('Test g mean: ', scores['test_gmean'].mean())

Test roc:  0.8951748528486864
Test accuracy:  0.8740143369175627
Test f1:  0.2044744575663871
Test g mean:  0.8139754143506067

from imblearn.pipeline import make_pipeline as make_imb_pipeline
oversample_pipe.fit(X_train,y_train)
prediction = oversample_pipe.predict(X_test)

from sklearn.metrics import confusion_matrix
import seaborn as sns

cf_matrix = cf_matrix = confusion_matrix(y_test,prediction)
sns.heatmap(cf_matrix/np.sum(cf_matrix), annot=True, fmt='.2%', cmap='Pastel1')

<matplotlib.axes._subplots.AxesSubplot at 0x7f4dc0485ad0>

from sklearn.metrics import ConfusionMatrixDisplay

ConfusionMatrixDisplay.from_predictions(y_test,prediction,cmap='viridis')

<sklearn.metrics._plot.confusion_matrix.ConfusionMatrixDisplay at 0x7f4dc0490150>

from sklearn.metrics import RocCurveDisplay

RocCurveDisplay.from_predictions(y_test,prediction)

<sklearn.metrics._plot.roc_curve.RocCurveDisplay at 0x7f4dc0410110>

from sklearn.metrics import roc_auc_score
from sklearn.metrics import accuracy_score
from sklearn.metrics import f1_score
from imblearn.metrics import geometric_mean_score

print('Roc score: ',roc_auc_score(y_test,prediction)) 
print('Accuracy score: ',  accuracy_score(y_test,prediction))
print('f1 score: ',f1_score(y_test,prediction))
print('G mean score: ',  geometric_mean_score(y_test,prediction))

Roc score:  0.8405286343612335
Accuracy score:  0.8790972595378829
f1 score:  0.24242424242424243
G mean score:  0.839550960322227

SMOTETomek

from imblearn.over_sampling import SMOTE
from imblearn.combine import SMOTETomek
over_undersample_pipe = make_imb_pipeline(SMOTETomek(random_state=42), LogisticRegression(max_iter=5000))
scores = cross_validate(over_undersample_pipe,
                        X_train, y_train, cv=5, scoring={'roc_auc':'roc_auc', 'average_precision':'average_precision','accuracy':'accuracy','f1':'f1','gmean':gm_scorer})
print('Test roc: ',scores['test_roc_auc'].mean()) 
print('Test accuracy: ', scores['test_accuracy'].mean())
print('Test f1: ', scores['test_f1'].mean())
print('Test g mean: ', scores['test_gmean'].mean())

Test roc:  0.8951748528486864
Test accuracy:  0.8740143369175627
Test f1:  0.2044744575663871
Test g mean:  0.8139754143506067

over_undersample_pipe.fit(X_train,y_train)
prediction = over_undersample_pipe.predict(X_test)

from sklearn.metrics import roc_auc_score
from sklearn.metrics import accuracy_score
from sklearn.metrics import f1_score
from imblearn.metrics import geometric_mean_score

print('Roc score: ',roc_auc_score(y_test,prediction)) 
print('Accuracy score: ',  accuracy_score(y_test,prediction))
print('f1 score: ',f1_score(y_test,prediction))
print('G mean score: ',  geometric_mean_score(y_test,prediction))

Roc score:  0.8405286343612335
Accuracy score:  0.8790972595378829
f1 score:  0.24242424242424243
G mean score:  0.839550960322227

from sklearn.metrics import confusion_matrix
import seaborn as sns

cf_matrix = cf_matrix = confusion_matrix(y_test,prediction)
sns.heatmap(cf_matrix/np.sum(cf_matrix), annot=True, fmt='.2%', cmap='Pastel1')

<matplotlib.axes._subplots.AxesSubplot at 0x7f4dc02bee10>

ConfusionMatrixDisplay.from_predictions(y_test,prediction,cmap='viridis')

<sklearn.metrics._plot.confusion_matrix.ConfusionMatrixDisplay at 0x7f4dc0405c90>

from sklearn.metrics import RocCurveDisplay

RocCurveDisplay.from_predictions(y_test,prediction)

<sklearn.metrics._plot.roc_curve.RocCurveDisplay at 0x7f4dc01df790>

from sklearn.decomposition import PCA
model_pca = PCA(n_components=12)

new_train = model_pca.fit_transform(X_train)
new_test = model_pca.fit_transform(X_test)

new_model = LogisticRegression()

new_model.fit(new_train, y_train)

LogisticRegression()

predict_train_pca = new_model.predict(new_train)

predict_test_pca = new_model.predict(new_test)

import pandas as pd
from sklearn.decomposition import PCA
from sklearn.linear_model import LinearRegression
from sklearn.metrics import mean_squared_error
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler

X = StandardScaler().fit_transform(X)
pca = PCA(n_components=.90)
pca.fit(X)
print("Variance explained by the principal components (in decrising order): ", pca.explained_variance_ratio_)
X1 = pca.transform(X)
print("Shape of transformed X : ", X1.shape)

Variance explained by the principal components (in decrising order):  [0.21764138 0.13485623 0.11982868 0.07645726 0.06841567 0.05096271
 0.04500212 0.04135184 0.04020125 0.04002362 0.0386388  0.03433675]
Shape of transformed X :  (7441, 12)

import plotly.express as px



pca = PCA(n_components=2)
components = pca.fit_transform(X)

fig = px.scatter(components, x=0, y=1, color=data['risksSEP'])
fig.show()

import matplotlib.pyplot as plt
import seaborn as sns

plt.figure(figsize=(15,10))
correlations = data.corr()
sns.heatmap(round(correlations,2),cmap='RdBu',annot=True,annot_kws={'size':7},vmin=-1,vmax=1)

<matplotlib.axes._subplots.AxesSubplot at 0x7f4dc00d8e90>

sns.clustermap(correlations,method='complete',cmap='RdBu',annot=True,annot_kws={'size':7},vmin=-1,vmax=1,figsize=(15,12))

<seaborn.matrix.ClusterGrid at 0x7f4dbee734d0>

sns.pairplot(data)

<seaborn.axisgrid.PairGrid at 0x7f4db740e590>

sns.pairplot(data, vars = ['Sex', 'TVdistance_m', 'p1_outdoorH'], hue = "risksSEP")

<seaborn.axisgrid.PairGrid at 0x7f4da474c210>

from imblearn.over_sampling import SMOTE
from imblearn.under_sampling import TomekLinks
over_undersample_pipe = make_imb_pipeline(SMOTE(random_state=42),TomekLinks(), LogisticRegression(max_iter=5000))
scores = cross_validate(over_undersample_pipe,
                        X_train, y_train, cv=5, scoring={'roc_auc':'roc_auc', 'average_precision':'average_precision','accuracy':'accuracy','f1':'f1','gmean':gm_scorer})
print('Test roc: ',scores['test_roc_auc'].mean()) 
print('Test accuracy: ', scores['test_accuracy'].mean())
print('Test f1: ', scores['test_f1'].mean())
print('Test g mean: ', scores['test_gmean'].mean())

Test roc:  0.8951748528486864
Test accuracy:  0.8740143369175627
Test f1:  0.2044744575663871
Test g mean:  0.8139754143506067

over_undersample_pipe.fit(X_train,y_train)
prediction = over_undersample_pipe.predict(X_test)

from sklearn.metrics import roc_auc_score
from sklearn.metrics import accuracy_score
from sklearn.metrics import f1_score
from imblearn.metrics import geometric_mean_score

print('Roc score: ',roc_auc_score(y_test,prediction)) 
print('Accuracy score: ',  accuracy_score(y_test,prediction))
print('f1 score: ',f1_score(y_test,prediction))
print('G mean score: ',  geometric_mean_score(y_test,prediction))

Roc score:  0.8405286343612335
Accuracy score:  0.8790972595378829
f1 score:  0.24242424242424243
G mean score:  0.839550960322227

from sklearn.metrics import confusion_matrix
import seaborn as sns

cf_matrix = cf_matrix = confusion_matrix(y_test,prediction)
sns.heatmap(cf_matrix/np.sum(cf_matrix), annot=True, fmt='.2%', cmap='Pastel1')

<matplotlib.axes._subplots.AxesSubplot at 0x7f4dc094add0>

ConfusionMatrixDisplay.from_predictions(y_test,prediction,cmap='viridis')

<sklearn.metrics._plot.confusion_matrix.ConfusionMatrixDisplay at 0x7f4dc0116190>

from sklearn.metrics import RocCurveDisplay

RocCurveDisplay.from_predictions(y_test,prediction)

<sklearn.metrics._plot.roc_curve.RocCurveDisplay at 0x7f4dc0006b10>